Restore full kinetic commit history (un-squash PR #112)#270
Merged
Conversation
…lcuation in DCON)
- Consolidated main program into a single `Main` function for better readability and modularity. - Introduced `PentrcControl`, `PentrcInternal`, and `PentrcOutput` structures to encapsulate control parameters, internal state, and output configurations respectively. - Created a new `Output.jl` file to manage output operations, including writing torque, orbit, and energy data to ASCII files. - Improved error handling and output directory management. - Updated documentation and comments for clarity and consistency.
- Updated module descriptions and added detailed function documentation for energy integration and torque calculations. - Refactored special functions to include elliptic integrals and double factorial computations. - Improved clarity and organization of code for better maintainability.
…d functions in Splines module
…management; implement spline_roots and spline_fit functions in CubicSpline
…RC into GeneralizedPerturbedEquilibrium Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
…States; add documentation page Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
…pline_roots with Roots.jl Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
…fix scoping, resolve duplicates - energy.jl: Replace TaskLocalValue with plain Dict (no TaskLocalValues.jl dep) - energy_integration.jl: Remove duplicate xintgrl_lsode stub (real impl in energy.jl), remove duplicate global Ref() definitions, add verbose keyword to tintgrl_grid - pentrc.jl: Add includes for energy_integration.jl and special.jl - Input.jl: Move `using DelimitedFiles` to top level, remove `using HDF5` from inside functions (already in pentrc.jl), remove shadowed local const declarations - Utils.jl: Remove dead Utilities sub-module with broken params.jl include Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- PentrcControl: @kwdef mutable struct with defaults for all fields, TOML-constructable via keyword splatting. Removed Fortran-era fields (idconfile, pentrc_threads, openmp_threads). - PentrcInternal: @kwdef mutable struct with fields for equilibrium-derived quantities (ro, bo, chi1, mthsurf, mfac) and profile interpolants (sq, kin, geom, dbob_m, divx_m) that will replace module globals. - PentrcOutput: Merged into PentrcControl (output config belongs with control parameters, matching ForceFreeStates/PerturbedEquilibrium pattern). - Output.jl: Updated to use PentrcControl + PentrcInternal instead of PentrcOutput. - Main.jl: Removed PentrcOutput usage and Fortran-era thread/clean logic. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
…through call chain - torque.jl: tpsi! now takes intr::PentrcInternal parameter; all bare globals (tdebug, dbob_m, divx_m, mthsurf, chi1, ro, bo, sq, kin, geom, mfac) replaced with intr.field references. calculate_fcgl updated similarly. - Compute.jl: compute_torque_all_methods! and compute_matrix_calculation! now take equil parameter. Removed references to deleted PentrcControl fields (data_dir, psi_out, wxyz_flag, output_netcdf). - Main.jl: Takes equil parameter, calls initialize_from_equilibrium! to populate PentrcInternal from PlasmaEquilibrium. - Input.jl: Added initialize_from_equilibrium!(intr, equil) function that extracts ro, bo, chi1, mthsurf from equilibrium. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
…line - Remove 12 dead test files referencing deleted dcon_interface.jl, pentrc_interface.jl, and SplineType (90 total stale references). - Wire PENTRC into GeneralizedPerturbedEquilibrium.main() pipeline: reads [PENTRC] section from gpec.toml, constructs PentrcControl via keyword splatting, initializes PentrcInternal from equilibrium. - Add commented [PENTRC] example section to DIIID-like gpec.toml. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
…ticForces Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
…ete stub energy_integration.jl Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
…n() only Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Neither file had any callers within the module. Elliptic integrals will be provided by Elliptic.jl when needed by future implementations. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
…int before FFT Fortran fspline_fit_2 (math/fspline.f:293) uses f = fst%fs(:, 0:my-1, iq), explicitly excluding the duplicated endpoint. Julia was FFT'ing all length(ys) samples including the θ=0 ≡ θ=2π duplicate, biasing the DC coefficient by ~(f(0) − mean)/N. Small effect (~1% for the a10 kinetic Bk matrix at ψ=0.30928), but principled — matches Fortran convention. Detects the duplicate automatically via ys[end] − ys[1] ≈ 2π; if ys is already non-duplicated, the FFT runs on all samples as before. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
…gral
The phase factor pl[i] = exp(-2πi·lnq·cum_wb_arr[i]/pl_denom) used a
left-Riemann cumulative sum of the bounce integrand, whereas Fortran
torque.F90:736-750 uses spline_fit + spline_int, which for smooth functions
produces a cubic-spline cumulative integral ≈ trapezoidal:
Fortran bspl%fsi(j)/Δx ≈ g_1 + g_2 + ... + g_{j-1} + g_j/2
Julia cum_wb_arr[i] = g_1 + g_2 + ... + g_{i-1} (too large by g_{i-1}/2)
The discrepancy was a per-sample phase offset of ~π·lnq·g_i/pl_denom, small
per-point but concentrating at the m=0 column of the kinetic matrices
because m=0 has no oscillating Fourier basis factor to average the
sampling bias out.
Effect at ψ=0.30928, a10 kinetic example (Bk matrix):
- Re(Bk[:, m=0]) Julia/Fortran ratio: 10.5 → 1.00
- Full Bk rel_frob: 4.30% → 1.60%
- Hk rel_frob: 30-100% → 1.61%
- Ck, Dk, Ek rel_frob: 2-5% → 0.5-1.9%
Fix: cum_wb_arr[i] = cum_wb − wb_integrand/2 (same for cum_wd_arr).
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
… integrals Convert the three right-Riemann sums in _bounce_integrate (bj_integral and the wmu_ba/wen_ba bounce-average loops) to explicit trapezoidal with 0.5 endpoint weights. The sums were formally right-Riemann over 2:ntheta, bit-equivalent to trapezoidal only because the payload arrays (jvtheta, wmu_mt, wen_mt) are zero at i=1 and i=ntheta from the 2:ntheta-1 population loop. Writing the weights explicitly makes the quadrature self-correct if the boundary handling ever changes. Also remove cum_wd_arr: allocated and written since introduction (163aef2) but never read. Mirrors a commented-out Fortran branch (torque.F90:747-751) for magnetic-precession-dominated regime that neither code activates. Numerical effect: all 6 kinetic matrix rel_frob values at a10 ψ=0.30928 unchanged (Ak 2.82%, Bk 1.60%, Ck 1.87%, Dk 0.54%, Ek 0.77%, Hk 1.62%). Tiny ~1e-6 residuals in kwmat Re and ktmat Im (noise from fwmm/ftmm resonance-operator symmetry) go to exact zero. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
…t/GPEC into feature/pe-kf-interface
…ic kwmat/ktmat Adds `integrate_pitch_gar_quadgk_wt` and `_pitch_gar_kernel_quadgk_wt!` that emit BOTH the fwmm half (rex=0, imx=1 → Fortran kwmat) and the ftmm half (rex=1, imx=0 → Fortran ktmat) in a single pitch integration, sharing one energy integration per (λ, E). Returns a length-`2*nqty` packed buffer `[wmm | tmm]`. The prior single-integration approach (rex=imx=1, then split kwmat=real, ktmat=imag) gives correct forward sums kwmat+ktmat = full but WRONG adjoint combinations kwmat-ktmat for non-Hermitian B_k, C_k, E_k blocks — because Fortran's kwmat and ktmat are each genuinely complex (matching Fortran pentrc/torque.F90:842-847 rex/imx assignments and pentrc/pitch.f90:376 integrand decomposition), not pure real / pure imaginary. Per-surface matrix dumps against Fortran dcon/fourfit.F:1080-1082 (`kwmat_l`, `ktmat_l`) confirm element-by-element agreement with the dual-output convention. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
…-triangle reconstruction Rewires `kinetic_energy_matrices_for_euler_lagrange!` and `compute_kinetic_matrices_at_psi!` to populate `kwmat` and `ktmat` directly via the new `integrate_pitch_gar_quadgk_wt` dual-output kernel, replacing the prior post-hoc `complex(real(full), 0)` / `complex(0, imag(full))` split. For A, D, H Hermitian-outer-product blocks (wz†wz, wx†wx, wy†wy) stored as upper triangles, the lower-triangle reconstruction now uses different mirrors per half: kwmat[j,i] = conj(kwmat[i,j]) (Hermitian) ktmat[j,i] = -conj(ktmat[i,j]) (anti-Hermitian) Derivation: S_w = complex(0, imag(xint)) is pure imaginary so conj(S_w) = -S_w, whereas S_t = complex(real(xint), 0) is pure real so conj(S_t) = S_t. Combined with conj(factor) = -factor (factor = -i/(2n)), these mirrors recover Fortran's independently-computed (j,i) slots exactly. Implemented via a `mirror_sign` parameter to a shared `_assemble_hermitian!` closure. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
…itian kinetic A
The previous implementation computed `aamat = (amat_lu \ amat_kin)'` which equals
`(A_kin⁻¹ · A_kin)' = I` exactly. This silently zeroed `umat_diff = I - aamat` and
DROPPED the `im·psio_over_n · umat_diff · b1mat` term from `paat`, the
`im·psio_over_n · aamat · bkmat` term from `r1mat`, and the
`im·psio_over_n · umat_diff · cmat_kin` term from `r2mat` whenever `amat_kin` was
non-Hermitian.
Fortran dcon/sing.f:1004-1008 computes
temp2 = amat_kin
zgbtrs("C", ..., amatlu, temp2) ! temp2 = A_kin⁻ᴴ · amat_kin
aamat = CONJG(TRANSPOSE(temp2)) ! aamat = amat_kin^H · A_kin⁻¹
which is NOT the identity when `amat_kin` is non-Hermitian. This was hidden while
the kinetic producer gave pure-real `kwmat` and pure-imag `ktmat` (amat_kin stayed
Hermitian); exposing it required the correct dual-output kernel (prior commits) that
makes amat_kin genuinely non-Hermitian via the anti-Hermitian ktmat contribution.
Fix: `aamat = (amat_lu' \ amat_kin)'` — the `amat_lu'` backslash path solves Aᴴ·x = b,
giving `temp = A_kin⁻ᴴ · amat_kin` which after adjoint becomes the Fortran form.
With all three kinetic fixes in place (dual-output kernel, anti-Hermitian ktmat
mirror, and this aamat correction), the DIIID_kinetic_example DCON benchmark
reaches Fortran agreement:
Re(et[1]): Julia 1.0481, Fortran 1.0507, err 0.25%
Im(et[1]): Julia -0.3012, Fortran -0.3013, err 0.01%
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
…ts_prim Clarifies that Julia stores the primitive (pre-Schur-reduction) geometric forms D = χ₁·(g23 + q·g33·m/n) and the analogous E. The `_prim` suffix follows the existing `fmats_prim` convention. Previously, a reader diffing against Fortran fourfit would find that Fortran saves two distinct splines — `dbats/ebats` (primitive) and `dmats/emats` (kinetic-block overwrites with χ₁²·(g22+2q·g23+q²·g33) and related) — whereas Julia's field named `dmats` actually holds the primitive form. The overwritten form is only consumed by Fortran's alternate on-demand singular-surface Schur path (`fkg_kmats_flag=.false.` in sing.f), which Julia does not implement. Renaming makes the intent explicit and avoids a silent naming trap. Pure rename; no numerical change. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
…rface # Conflicts: # src/Equilibrium/DirectEquilibrium.jl
…lytic resonance poles Port the QuadGK energy-integration robustness from PR #220 into the KineticForces energy integrator, with a corrected pole decomposition. - Map x = E/T to u = 1-exp(-x) on [0,1), absorbing the Maxwellian weight into du, so the integral covers [0,infinity) without an xmax cutoff. - Remove each resonance pole analytically by a Sokhotski-Plemelj decomposition: subtract R/(u-u_pole) and add back the elementary integral R*(log(1-u_pole)-log(-u_pole)). The same complex pole u_pole (x_pole = x_res - i*nu/Omega') is used in both the subtraction and the add-back, so the decomposition is exact and converges for nu > 0 as well as nu -> 0 (PR #220 subtracted at the real u_res but added at the complex u_pole, which diverges for collisional cases). - Collisionless limit uses the causal Sokhotski-Plemelj branch. - CGL has no resonance denominator: integrate the physical x-space integrand directly over [0,infinity) via QuadGK. - ximag is accepted for signature compatibility but is now unused. Adds find_resonance_energies and unit tests: resonance-root cases, a collisional cross-check against direct x-space quadrature, and a collisionless-limit continuity check. 110 kinetic unit tests pass. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
…e resonances When kinetic_source="calculated" runs the full KF torque pipeline, the pitch loop can hand the energy integrator parameter sets whose resonance polynomial has a root at very large or very small x. Two corresponding NaN traps appear in the Sokhotski-Plemelj analytic pole contribution: - x_res > 700: exp(-x_res) underflows to zero, R becomes 0+0im, and R*(log(1-u_pole) - log(-u_pole)) evaluates as 0*(-Inf+iπ) = NaN. The resonance's true contribution is identically zero (the Maxwellian weight is below machine precision) — skip the resonance. - nu_res / omega_prime = Inf (harmonic ν overflows at tiny x_res, or omega_prime ≈ 1e-30): exp(-x_pole) at infinite imaginary argument returns NaN. The resonance is infinitely collisionally broadened — no localized pole, no extraction needed; skip and let QuadGK integrate the smooth integrand directly. Verified by the regression-harness solovev_kinetic_calculated case (which exercises the calculated-source KF pipeline that the "fixed" fullruns do not): pre-port et[1] = 19.914 - 0.628i, post-fix et[1] = 19.908 - 0.658i (0.03% / 5% relative). 110 kinetic unit tests still pass. Also updated: - test/test_data/regression_solovev_kinetic_calculated/gpec.toml: replace removed dmlim/set_psilim_via_dmlim keys with develop's psiedge edge-scan band so the calculated-source case is runnable. - test/runtests_fullruns.jl: refresh ex4 (Solovev kinetic multi-n) baseline to the develop-consistent eigenvalue 0.22325 — the prior -0.01248 dates from before develop's edge-scan and periodic-theta-endpoint evolution. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
…schema The DIIID kinetic benchmark's auto-generated gpec.toml used the deprecated set_psilim_via_dmlim and dmlim keys, which develop removed (replaced with the psiedge edge-scan band). The new FFS rejects unknown keys, so the benchmark errored out before any compute. Drop the deprecated pair; the existing psiedge = 1.00 (no edge-scan truncation) is the closest current equivalent. A/B verified with the new u-substitution energy integrator on the merged tree: same DIIID benchmark numbers as the pre-port integrator (FGAR T = 1.805 N·m vs Fortran 1.780, 1.39% err; FGAR dW = 0.160 vs 0.137, 16.94% err). The dW drift from PR #224's body (0.141 → 0.160) is caused by the develop merge upstream of KF (most likely the periodic-theta-endpoint fix changing bounce-data evaluation), not by this port — the energy integrator is faithful. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
KineticForces — Complete PE→KF pipeline and validate DIIID-PENTRC benchmark
…o current FFS schema The dmlim / set_psilim_via_dmlim edge-scan mechanism was removed in develop; the auto-generated gpec.toml in benchmark_diiid_kinetic_dcon.jl still carried those keys and would fail at config parse. Drop them and keep psiedge = 1.0, mirroring the analogous fix already applied to benchmark_diiid_kinetic.jl in commit ace7abc. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
…Julia 1.12
The powspace 'basic properties' test checked
@test all(diff(pts) .> -eps(1.0))
which permits at most 1 ulp of non-monotonicity per step. The
_powspace_antideriv polynomial for pow=9 evaluates ~20 terms with mixed
signs near |x|=1, so a few ulps of FP rounding noise is normal. CI Julia
1.12 (vs 1.11) reorders the FMA chains in the @. broadcast and crosses
the 1-ulp threshold on one consecutive pair (Julia 1.11 passes; 1.12
fails). Loosen the tolerance to 100*eps(1.0) ≈ 2.2e-14 — still 10⁻⁸×
the typical 0.01 step size, so any real algorithmic non-monotonicity
bug is still caught.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
Pre-merge sweep on PR #112 surfaced three actionable items beyond the green CI; this commit addresses them in a single pass. Test coverage for the two NaN-prevention guards added in 382c31e to src/KineticForces/EnergyIntegration.jl: * xr > 700 Maxwellian-underflow skip (line 207): constructed via the linear-wd case with we=-30 to place x_res at 900. Without the guard the residue R=0 multiplied by log(-u_pole)→-Inf trips the 0·∞ NaN trap. New @testset asserts the result is finite. * isfinite(pole_offset) overflow skip (line 220): constructed via tiny wb=1e-15 (omega_prime ≈ 5e-16) with Krook nuk=1e300 to force nu_res/omega_prime ≈ 2e315 → Inf. Parameters stay above SINGULAR_EPS so find_resonance_energies still returns the resonance. New @testset asserts the result is finite. Both gaps were documented in the energy-integrator NaN-guards memory as "failure modes only surface end-to-end, not in unit tests." Now they do. Coverage for the exported diagnostic evaluate_energy_integrand: previously only the scalar form energy_integrand_scalar was tested. New @testset checks return type, length, finiteness, and element-wise agreement with the scalar form (the broadcast it documents itself as). Cleanup: promote the bare 1e-30 literal to module-level constant SINGULAR_EPS in KineticForces.jl, used at six sites across EnergyIntegration.jl (resonance solver + Sokhotski-Plemelj denominator) and BounceAveraging.jl (Pfirsch-Schlüter denominator regulariser). The name documents the intent (treat-as-zero / singularity guard) instead of leaving readers to infer it. Docstring added for _bisect_vpar in BounceAveraging.jl (line 361). Documents the bracket-sign assumption, what the tol and maxiter knobs control, and the fallback when iteration is capped. Full test suite: 892/892 pass locally including +10 new kinetic tests (KineticForces unit tests 110 → 120). Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
…tained validations Trim kinetic-PR benchmark noise: keep only the two benchmarks that validate distinct, durable physics outputs against Fortran, and remove bug-hunt aids. Deleted: - benchmark_a10_kinetic_dcon.jl (fast-iteration dev-loop duplicate) - compare_kinetic_matrices.jl + dump_julia_kinetic_matrices.jl (2-file matrix dump/compare dance; manual pentrc.in edit, l=0 only) - _plot_cond_fbar.jl (shim superseded by Analysis.ForceFreeStates.plot_cond_fbar) Renamed for clarity: - benchmark_diiid_kinetic.jl -> benchmark_diiid_ideal_ntv_torque.jl - benchmark_diiid_kinetic_dcon.jl -> benchmark_diiid_kinetic_stability.jl The stability benchmark now calls Analysis.ForceFreeStates.plot_cond_fbar directly instead of the removed shim. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
…ted-source kinetic-MHD regression Remove six unreferenced synthetic fixtures under test/data/ (idcon_test.bin, peq_*.txt, pmodb_*.txt) seeded during the preliminary PENTRC->Julia conversion; no test or source reader references them. Re-enable the self-consistent kinetic-MHD FFS regression (kinetic_source= "calculated") on the Solovev case, now that the Julia kinetic-DCON path is validated against Fortran kinetic DCON (<2% on Re/Im of et[1], DIIID, PR #112). Pin the actual HEAD baseline et[1] = 15.888 - 0.711 i: Re tight (well- conditioned), Im bracketed loosely (kinetic damping, FP-sensitive). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
…r θ=2π endpoint-trim fix The kinetic-branch FourierCoefficients fix (7b956d7) drops the duplicated θ=2π endpoint before the metric FFT, matching Fortran fspline_fit_2 (equil/fspline.f:293). The pre-fix FFT double-counted θ=0, biasing the metric DC coefficient, so every Fourier-derived quantity shifts slightly. This is the root cause of the PR #112 CI failure in the STRIDE BVP Δ′ regression. - Remove the fft_flag control: it was declared but never used (make_metric always took the FFT path). In Fortran idcon.f:619 fft_flag selects fspline_fit_2 (FFT, trims endpoint) vs fspline_fit_1 (integration DFT, keeps endpoint). Julia only implements the FFT path, so commit to it and drop the dead flag rather than gate behavior on a flag that does nothing. - Re-pin the STRIDE BVP Δ′ regression (runtests_parallel_integration.jl): real parts dpm[1,1] 8.357→8.673, dpm[2,2] -3.995→-3.891. The old values were the buggy double-counted FFT; the new values faithfully reproduce fspline_fit_2. - solovev_kinetic_multi_n et[1] (runtests_fullruns.jl): the trim shifts this documented ill-conditioned near-marginal eigenvalue across zero (-0.19 → +0.19). Replace the sign assertion with a sign-agnostic magnitude bracket; the well-conditioned et[2]/et[3] remain pinned tightly as the real regression guards. - Fix stale comment path (math/fspline.f → equil/fspline.f). Full test suite green locally (all files, including those CI never reached before because the Δ′ failure halted the include chain). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
…) bracket to 8% for θ=2π trim + Julia-version spread CI runtests 1.11 (Linux x86) computes Im(et[1]) = -0.7548 for the self-consistent calculated-kinetic Solovev anchor, 6.2% below the -0.711 pin (rtol was 5%), while runtests 1.x and macOS land within 5% on the high side. The θ=2π endpoint-trim correctness fix plus Julia-version FP reassociation spread this loosely-bracketed damping rate across -0.675 … -0.755. Keep the -0.711 physics anchor and widen the already-loose Im bracket from 5% to 8% (range -0.768 … -0.654, covers all platforms). Re(et[1]) stays well within its tight 1% guard, so it remains the real regression check. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
… from DIIID example toml The commented-out [KineticForces] stub (fgar_flag, verbose) is a placeholder that cannot run without kinetic profiles for the DIII-D-like example. Removing it to avoid implying the section is usable as-is; re-adding it is tracked by issue #268 (provide kinetic profiles for the DIII-D-like example to enable an NTV-torque demonstration). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
This reverts commit b72ddc8.
PR #112 was squash-merged into develop as b72ddc8, collapsing all 96 kinetic commits into one. This restores the full commit history: the preceding commit reverts the squash, and this --no-ff merge brings origin/kinetic back with its complete ancestry plus the two post-merge follow-ups (toml cleanup, comment trim). Net content is identical to the squashed result.
d-burg
added a commit
that referenced
this pull request
Jun 10, 2026
Brings develop's post-perf/riccati work (power-norm eigenvalues #225, kinetic-history restore #270/#112, KineticForces/NTV module, GridUtilities, vacuum/* → FreeBoundaryStability/* + SurfaceGeometries/* HDF5 reorg, faithful Fortran θ=2π endpoint drop before the metric FFT) onto the tearing branch. Conflict resolutions (5): - src/Utilities/Utilities.jl: include-union — keep tearing's PhysicalConstants/KineticProfiles/NeoclassicalResistivity AND develop's GridUtilities (the latter injects powspace directly, no submodule). - test/runtests.jl: test-union — tearing's SLAYER/dispersion/kinetic_profiles/ resist suites AND develop's runtests_kinetic.jl. - test/runtests_fullruns.jl: took develop's sign-AGNOSTIC kinetic multi-n pin (abs(real(et[1])) < 0.5). Supersedes the tearing bracket: develop's endpoint drop shifts the near-marginal et[1] across zero (+0.190 macOS), so real(et[1])<0 no longer holds. Well-conditioned et[2]/et[3] tight pins retained. - src/GeneralizedPerturbedEquilibrium.jl (main): kept BOTH independent post-PE stages — develop's KineticForces (NTV) block and tearing's SLAYER stage (_run_slayer_stage closure, runs under force_termination too). - src/GeneralizedPerturbedEquilibrium.jl (HDF5 writer): adopted develop's new output schema (FreeBoundaryStability/*, FreeBoundaryStability/XiNorm/*, SurfaceGeometries/*, singular/kinetic/*) and dropped the superseded vacuum/* keys; kept tearing's singular/delta_prime_raw output. Verified: package instantiates + precompiles; Tearing, KineticForces, GridUtilities, NeoclassicalResistivity all present. No code reads the dropped vacuum/* keys (one stale docstring ref in ForceFreeStatesStructs remains, inherited from develop). Full suite validation pending. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
3 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Squashing this PR would re-collapse the history and defeat its entire purpose.
What this does
PR #112 was squash-merged into develop as
b72ddc8e, collapsing all 96 kinetic commits into a single commit and losing the per-commit history. All 96 commits remained intact onorigin/kinetic, so this PR restores them into develop's graph without rewriting any public history (no force-push).Two commits:
ec5aa4c7—Revert "KineticForces Module (formerly PENTRC) (#112)"(undoes the squashed content)93ca05a0—--no-ffmerge oforigin/kinetic, bringing the full 96-commit ancestry back, plus the two post-merge follow-ups that never made it into the squash:494b7c2bEXAMPLES - CLEANUP - Drop placeholder commented[KineticForces]block (tracked by Provide kinetic profiles for the DIII-D-like example to enable an NTV-torque demonstration #268)b95d018cREGRESSION - MINOR - reduces comment lengthNet effect
examples/DIIID-like_ideal_example/gpec.tomlandtest/runtests_fullruns.jl(8 deletions).7b956d71,47cb6f4a, early75946c27are all ancestors of the merge).b72ddc8e(squash) → revert → this merge. That squash→revert noise is the accepted cost of un-squashing without a force-push.🤖 Generated with Claude Code